JPS60218700A - How to understand signal limits - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention] rB[1)-(7) 禾１１田4) II 1 The present invention includes
As a discrimination criterion between the signal of interest and the background signal or background signal mixture, in order to understand the signal limitations that arise before background signal mixing, in particular the signal limitations for speech processing of spoken words before background noise. Concerning the way in which their amplitude behavior is exploited.

[Conventional technology]

Various methods are known in the art for word limit detection, ranging from simple analysis of signal amplitudes to analysis of time signals, with subsequent recognition and classification of pauses and speech.

However, evaluation of the average amplitude or amplitude behavior of the background signal is generally not sufficient for many applications. For example, as is known from the prior art, a complete analysis of parametric frequency and/or amplitude behavior is expensive in many applications, both in software and hardware.

[Problem that the invention seeks to solve]

It is an object of the present invention to provide a method of the type mentioned in the opening, which is inexpensive in terms of hardware and software, operates relatively accurately, and which also eliminates certain unimportant signal disturbances (e.g. door knocks, street noises). , the noise of many people, etc.).

Means for Solving Problem C] According to the present invention, this object is achieved by the method described in claim 1.

Preferred embodiments of the present invention are shown in the following claims.

〔Example〕

The invention will be explained in more detail below by means of embodiments shown in the drawings.

As mentioned above, the method according to the invention uses the amplitude behavior of the signal of interest and the background signal or background signal mixture as a criterion for discrimination. To this end, in a first step, an input signal is received and subsequently a preprocessed signal or signal mixture, ie an input quantity E1 whose average amplitude A and its 1
observed with respect to a first passage frequency N with reference to one reference quantity R and its variation range dN within a given time interval, in which case the passage frequency N is averaged from the individual passage frequencies, and in that case The fluctuation range is determined by the maximum deviation of the individual passing frequencies. For this purpose, the input signal or signal mixture Z (t) is first amplified, then filtered by a bandpass filter and subsequently subjected to an analog-to-digital conversion, resulting in the above-mentioned input quantity E ( (See Figure 1). From the obtained amount, the supplementary amount SiN
2 is derived (see Figure 3).

In a second step, the actual passing frequency N1 is determined with reference to the reference quantity R during the time slots required for sufficient resolution. Fluctuation range dN and actual passing frequency N
1, the previously derived auxiliary quantity S1 or N2 is the evaluation quantity S
can be mapped to Based on this evaluation amount S, the actual input i1
B is measured. For this purpose, an operation O1 is performed, which is related to the position of the input quantity E relative to the evaluation quantity S.

Two limit values UG, O based on the type of signal of interest
GI is determined. The result of operation 01 is the first limit value UG
is limited downward by. The presence of an interesting signal is recognized when the second upper limit value OGl is reached. The exact signal start SB is then located before the relevant recognition instant ZEIO by a certain time fragment (see FIGS. 4 and 5).

In a third step, another operation 02 causes the position of the input quantity E relative to the evaluation quantity S to reach a second limit value OG2 previously determined on the basis of the type of signal of interest. The absence of the signal of interest detected in the second stage is then confirmed. The exact signal end SE lies before the relevant recognition time ZE2 by a certain time fragment (see FIGS. 6 and 7).

According to a preferred embodiment of the invention, said operation 01 is an integral process.

The exact signal start SB is determined by the temporal position of the final value of the integration result, equal to the lower limit value UG and before the relevant recognition time ZEI.

If the input address E exceeds a threshold adapted to the background signal mixture, it is advantageous that the first step as described above is repeated. Furthermore, it is expedient that a waiting time is inserted between the first and second stages.

The above auxiliary quantity is 5l=a*A+c S2=b*A+c N2=N/d where the quantity a % b % C,,' d is a constant determined by experience or by the type of signal of interest. It is defined as.

The evaluation amount is S= 32...(N2~dN)<Nl<(N2+d
In the case of N), Sl... is defined as all other cases.

The first operation is 01=Σ5IGN('IEI-5) It is defined as.

The second phonetic calculation is 02 = 02 + 1. If IEI<S, then 0... in all other cases or OG2... in all other cases, 02-1...
It is defined as a case where IEI<.

Although the processing and evaluation steps according to the invention can be carried out by digital circuits, it is also expedient to carry them out by means of a microprocessor and a corresponding program.

FIG. 2 shows a block diagram for word limit detection. In this block diagram, the aforementioned stages are again clearly shown: the first stage, the second stage and the third stage.

FIG. 8 is an explanatory diagram of the full word limit recognition of the spoken word "5top", and the upper diagram shows the oscillation course of the relevant time signal with associated upper and lower thresholds. , the middle diagram shows the digital indication signal generated in the state "signal of interest present", and the lower diagram shows the course of the recognition of word start and word end.

According to the method of the invention, the following advantages are obtained in connection with the recognition of the limits of the spoken word: 1. The signal with the frequency behavior of the ambient noise is compared with a high threshold and therefore has a large amplitude. Even in rare cases, an erroneous "word start notification" can be detected.

2. Signals with other frequency behavior can be recognized even if their amplitude is small, since they are compared with a lower threshold.

3. For signals whose main part does not exceed the respective threshold, the integral of the "clipped" value cannot reach a certain predetermined absolute value. 'Then there was a knock on the door,
Weak signals such as manufacturing machine noise or speech or music from a distance are suppressed.

4. Even speech signals with weak starting fricatives can be detected correctly by the integration method.

In most cases, ambient noise has predominant frequency components within the vowel formants. However, since these formants often have relatively large amplitudes, they can be detected even at high thresholds.

In contrast, unvoiced noises with large components at high frequencies and which are strongly attenuated by pre-processing are usually distinguished from the ambient noise only by their zero-pass rate. This is taken into account by choosing a low threshold.

The method according to the invention is of course not restricted to the embodiments described above. For example, it can also be used for monitoring purposes, for example radio monitoring purposes, to find out certain typical signal profiles within a signal mixture.

[Brief explanation of drawings]

1 is a schematic block diagram illustrating the basic principle of preprocessing of the time signal of an input signal applied to the input terminal of a device for carrying out the method according to the invention; FIG. a block diagram of the steps; FIG. 3 is a schematic block diagram illustrating the apparatus and their cooperation for the implementation of the first method step; FIG. 4 is a schematic block diagram showing the apparatus and their cooperation for the implementation of "start detection" Block diagram, FIG. 5 is an illustration of the example of the spoken word "5top", FIG. 6 is a block diagram showing the device for implementation of "end detection" and its cooperation, FIG. 7 is the spoken word "5top" An explanatory diagram of an example of “end detection” of the word “5top”, Figure 8 shows the spoken word “5top”
FIG. 2 is an explanatory diagram of an example of full word limit recognition. A...Average amplitude, dN...Fluctuation range, E...Input amount, N, Nl...Passing frequency, N2...Auxiliary amount, Of
02...Calculation, OGI, OG2...Limit value, R...
・Reference amount, S...Evaluation amount, Sl, S2...Support amount,
SB...signal start, SE...signal end, UG...
Limit value, ZEl, ZE2...point of recognition. leprosy

Claims (1)

[Claims] 1) The signal of interest and the background signal or background in order to understand the signal limits that arise before background signal mixing, especially the signal limits for speech processing of words spoken before background noise. In a method in which their amplitude behavior is used as a discrimination criterion between signal mixtures, in a first step the acceptor is accepted and subsequently the preprocessed signal or signal mixture, i.e. the input quantity (E
), is observed with respect to its average amplitude (A) and its first passing frequency (N) with respect to its one reference quantity (R) and its variation range (dN) within a given time interval, and its In this case, the passing frequency (N) is averaged from the individual passing frequencies, and the fluctuation range is determined by the maximum deviation of the individual passing frequencies, and the supplementary amounts (SL, S2, N
2) is derived, and in a second step the actual passing frequency (Nl) is determined during the time slots required for sufficient resolution by the reference quantity (R
), with reference to the observed and derived passage frequency (N2) with a fluctuation range (dN) and the actual passage frequency (N1), the previously derived supplementary quantity (31
or S2) is associated with the evaluation amount (S), the actual input amount (E) is measured by this evaluation amount (S), and the position of the input amount (E) relative to the evaluation amount (S). An operation (01) related to is performed, two limit values (UG, 0G1) are determined based on the type of signal of interest, and the result of operation (01) is lowered by the first limit value (UG). the presence of the signal of interest is recognized upon reaching the second upper limit value (OGI), and the exact signal onset (S In a third step, another operation (02) determines the position of the input quantity (E) relative to the evaluation quantity (S), based on the type of signal of interest. The absence of the signal of interest detected in the second stage is confirmed upon reaching the determined second limit value (OG2-), and the precise signal termination (SE) is determined to be relevant only for a specific time fragment. A method for grasping the limit of a signal, characterized in that the signal is located before the recognition point (ZE2). 2) The method according to claim 1, wherein the operation (01) is an integral process. 3) The exact signal start (SB) is determined by the temporal position of the final value of the integration result equal to the lower limit value (UG) and before the relevant recognition instant (ZEI). The method described in Scope No. 2. 4) A method according to claim 1, characterized in that the pre-processing includes layer width, bandpass filtering and absolute value formation. 5) A method according to claim 1, characterized in that the pre-processing comprises an analog-to-digital conversion. 6) Method according to claim 1, characterized in that the first step is repeated if the input address (E) exceeds a threshold adapted to the background signal mixture. 7) A method according to claim 1, characterized in that a waiting time is inserted between the first stage and the second stage. 8) The auxiliary quantity is 51=a*A+c S 2- b * A + c N2=N/d where the quantities a, b, c, d are determined by experience or by the type of signal of interest. The method according to claim 1, characterized in that the constant value is defined as a constant. 9) The evaluation amount is S= 3.2...(N2-dN)<Nl< (,N,
2+dN) ゛Sl... in all other cases. The method according to claim 1. 10) A method according to claim 1, characterized in that the first operation is defined as 01-ΣS IGN (I E I -3). 11) If the second operation is 02 = 02 + 1, IEI < 3, then 0... in all other cases, or OG2... in all other cases, 02-1...
2. A method according to claim 1, characterized in that the IE is defined as a single case. 12) A method according to claim 1, characterized in that the processing and evaluation steps are carried out by digital circuitry. 13) A method according to claim 1, characterized in that the processing and evaluation steps are carried out by a microprocessor.